otx.algo.object_detection_3d.backbones.monodetr_resnet#

MonoDetr backbone implementations.

Classes

Backbone(name, train_backbone, ...)

ResNet backbone with frozen BatchNorm.

BackboneBase(backbone, train_backbone, ...)

BackboneBase module.

BackboneBuilder(model_name)

DepthAwareTransformerBuilder.

Joiner(backbone, position_embedding)

Joiner module.
class otx.algo.object_detection_3d.backbones.monodetr_resnet.Backbone(name: str, train_backbone: bool, return_interm_layers: bool, dilation: bool, **kwargs)[source]#

Bases: BackboneBase

ResNet backbone with frozen BatchNorm.

Initializes Backbone module.

class otx.algo.object_detection_3d.backbones.monodetr_resnet.BackboneBase(backbone: Module, train_backbone: bool, return_interm_layers: bool)[source]#

Bases: Module

BackboneBase module.

Initializes BackboneBase module.

forward(images: Tensor) → dict[str, NestedTensor][source]#

Forward pass of the BackboneBase module.

Parameters:

images (torch.Tensor) – Input images.

Returns:

Output tensors.

Return type:

dict[str, NestedTensor]

class otx.algo.object_detection_3d.backbones.monodetr_resnet.BackboneBuilder(model_name: str)[source]#

Bases: object

DepthAwareTransformerBuilder.

Constructor for Backbone MonoDetr.

class otx.algo.object_detection_3d.backbones.monodetr_resnet.Joiner(backbone: Module, position_embedding: PositionEmbeddingSine)[source]#

Bases: Sequential

Joiner module.

Initialize the Joiner module.

Parameters:

  • backbone (nn.Module) – The backbone module.

  • position_embedding (PositionEmbeddingSine) – The position embedding module.

forward(images: Tensor) → tuple[list[NestedTensor], list[Tensor]][source]#

Forward pass of the Joiner module.

Parameters:

images (torch.Tensor) – Input images.

Returns:

Output tensors and position embeddings.

Return type:

tuple[List[NestedTensor], List[torch.Tensor]]